One system for the whole school. Marking first.

One application. The roles you hold decide what you see. A teacher sees teaching tools; a SENCO sees SEND tools; a business manager sees the office work. Change someone's role and their screen changes, with nothing reinstalled.

Teacher Manager is one system for the adults in a school. Marking and feedback lead, because that is the work that eats the evenings. The office paperwork, planning and resources, leadership admin and the tutoring module sit in the same system, sharing the same models and the same records.

None of it is handed to someone else's cloud. One system to learn, one place the data lives, one set of permissions to govern.

The phases below are what we are working towards, not delivery guarantees.

Phase 1

The whole suite, working end to end, in a pilot school's hands. Not a demo of one feature: the complete loop, from a stack of exercise books to approved feedback, with every module present.

The models are trained and evaluated. What we are building now is the integration into the tools teachers actually use.

Designed

— specified and scoped, not yet built.

Model trained

— the model exists and has been evaluated; we are building it into the interface.

In integration

— being built into the tool a teacher actually uses.

Integrated

— in the product and working.

Phase 2

Depth where the founding schools tell us it is thin. The SENCO pack in full — IEPs, EHCP contributions, inclusion strategies, Pupil Premium statements. More subjects and question types, including working and notation in maths and science. Trust-level views that never expose a named pupil. And the evidence only a pilot can produce: marking agreement measured against your own markers, and the time actually saved.

Phase 3

The loop closes. The tutor module runs with pupils in the pilot schools, targeting the gaps marking revealed, and the results come back to the teacher. Marking to intervention to evidence, inside one system, with the data staying in the building. We don't know of another tool that does this, because it needs both ends.

Own the model. Keep the data. Inspect the reasoning.

We publish what we have measured, and we say plainly what is still being measured. Nothing here is an independent certification: the system is built to conform to the DfE's product safety expectations, and independent review is the step we are committing to in public.

• Does the marking agree with the mark scheme?

  We are running that measurement now, on real UK curriculum mark schemes, and we will not put a number up until people have validated it.

  What this means for your school: You will not be asked to trust a marketing figure. The teacher stays the marker of record and reviews every mark, and when we can show how closely it matches a real mark scheme, you will see the number and how we got it.

• Every answer is grounded in a source.

  A marking decision points back to the material it came from, and the reasoning is visible to the teacher who signs it off.

  What this means for your school: A teacher never has to take the tool's word for it. They can see why a mark was given and overrule it in seconds. No black box in a child's record.

• We test our own models for privacy leakage.

  Every model is tested with a published attack from the research literature, to check whether an attacker could tell what it was trained on. Anything that shows a signal does not ship as it is. That is a standing policy, not a one-off.

  What this means for your school: It is not only that pupils' work stays in the building. The models themselves are checked so they cannot quietly memorise and leak a child's data. It is the difference between promising privacy and testing for it.

• Trained on UK national infrastructure.

  Dawn, at the University of Cambridge, accessed via AIRR.

  What this means for your school: The model your pupils' work meets was built in the UK, on national research infrastructure, under UK law, not on a US cloud account.

  Read the write-up: how the training run went and what it cost.

Everything we claim about data, safeguarding and governance is documented in the open.

Last reviewed 23 September 2026.